WTF Summit Recap: The AI Revolution Two Years In—Reality Versus Hype

It's the early days of the artificial intelligence revolution. At this stage, enterprises are developing mostly tactical AI use cases. The next step is a more strategic, higher-value approach.

As part of The Information’s San Francisco AI Summit, Stephanie Palazzolo, The Information’s AI reporter and author of the AI Agenda newsletter, sat down with three AI executives to discuss how they see the AI revolution progressing:

• Lenore Lang, executive vice president of technology, media and telecommunications, Salesforce 
• Paige Costello, head of artificial intelligence, Asana
• Mikey Kindler, co-founder and chief operating officer, Lighty AI 

Making AI Core to the Business

AI is seen as suited for specific tasks such as customer service, document review or software engineering. The way forward is by experimenting with use cases outside these initial buckets and embedding AI deeper into the core of the business.

Lighty’s Kindler pointed out that one of the biggest misconceptions about AI, especially among enterprise customers, is that it is separate from the rest of the business. She believes the right approach is to embrace AI end to end across a business. “I think the AI is ready for it,” she said.

Asana’s Costello noted, and Kindler agreed, that reaching this next level of AI implementation requires redesigning business processes and workflows to embed AI in them. That may require slowing things down from taking a more experimental approach to focus on defining the biggest business problems and how to use AI to solve these problems more intelligently. “How we think about AI really changes the value we get from it,” said Costello.

Setting AI Up for Success

Lack of trust in AI is a major roadblock. The accuracy of machine-generated outcomes is often in doubt, which is why many AI use cases are relegated to the least sensitive business areas and AI gets the short end of the stick when it comes to trust. “Users are far more critical of AI than they are of humans,” says Kindler. She notes that when her human assistant made a mistake by scheduling a meeting with the wrong “John,” nobody thought much of it. But when AI makes a similar mistake, “it’s ‘The sky is falling.’”

To build trust in AI, says Costello, it’s necessary to reduce the probability of errors and increase detectability. The former boils down to giving AI the right amount of agency, while the latter is about designing models more transparently, to enable tracing and understanding AI’s decision-making.

“We need to work with AI to help itself,” said Costello. That means analyzing what AI did and why, and then giving the feedback to the person who set up the workflow using AI about how to improve it. “At the end of the day, no matter what AI does, you are responsible for the decisions,” said Costello.

Paying for AI

Some of the recently released AI models are expensive, and it’s sometimes hard to predict how much companies will use them. That means costs can explode unexpectedly, noted The Information’s Palazzolo, who asked panelists how they see the pricing for running AI models evolving.

Salesforce’s Lang said the company’s agreements with enterprises looked very different 10 years ago from what they will look like in a year’s time. Having traditionally relied on a per-user pricing model, Salesforce is moving to a consumption-based model. “You will find more creative pricing than you’ve ever seen before from us,” including bundles and individual packages, said Lang.

Lenore said that Salesforce is open to “whatever works best for a customer.” But she noted, “Hyperscalers have taught consumers and enterprises how they want to procure, and we’re going to model off of that moving forward.”

Designing for Tomorrow’s Models

When building products, there is thus a natural tension between anticipating the new generation of models and building for the models that exist today.

“We are not designing for the models that exist right now,” said Kindler. “We need to build whatever we think is best for our end user. And then we’ll figure the other stuff out on the back end. We’re not in the weeds feeling limited by what current technology can do.”

Asana’s Costello agreed that product development is driven by customer needs: “In designing AI tools, we’re all about who’s doing what on the Asana platform, why, and when the project needs to be done.”